Study on Delaunay Tessellations of 1-Irregular Cuboids for 3D Mixed Element Meshes
Total Page:16
File Type:pdf, Size:1020Kb
Study on Delaunay tessellations of 1-irregular cuboids for 3D mixed element meshes David Contreras and Nancy Hitschfeld-Kahler Department of Computer Science, FCFM, University of Chile, Chile E-mails: dcontrer,[email protected] Abstract Mixed elements meshes based on the modified octree approach con- tain several co-spherical point configurations. While generating Delaunay tessellations to be used together with the finite volume method, it is not necessary to partition them into tetrahedra; co-spherical elements can be used as final elements. This paper presents a study of all co-spherical elements that appear while tessellating a 1-irregular cuboid (cuboid with at most one Steiner point on its edges) with different aspect ratio. Steiner points can be located at any position between the edge endpoints. When Steiner points are located at edge midpoints, 24 co-spherical elements appear while tessellating 1-irregular cubes. By inserting internal faces and edges to these new elements, this numberp is reduced to 13. When 1-irregular cuboids with aspect ratio equal to 2 are tessellated, 10 co- spherical elementsp are required. If 1-irregular cuboids have aspect ratio between 1 and 2, all the tessellations are adequate for the finite volume method. When Steiner points are located at any position, the study was done for a specific Steiner point distribution on a cube. 38 co-spherical elements were required to tessellate all the generated 1-irregular cubes. Statistics about the impact of each new element in the tessellations of 1-irregular cuboids are also included. This study was done by develop- ing an algorithm that construct Delaunay tessellations by starting from a Delaunay tetrahedral mesh built by Qhull. arXiv:1312.1181v1 [cs.CG] 4 Dec 2013 1 Introduction Scientific and engineering problems are usually modeled by a set of partial dif- ferential equations and the solution to these partial differential equations is calculated through the use of numerical methods. In order to get good results, the object being modeled (domain) must be discretized in a proper way respect- ing the requirements imposed by the used numerical method. The discretization (mesh) is usually composed of simple cells (basic elements) that must represent the domain in the best possible way. In particular, we are interested in meshes for the finite volume method [1] which are formed by polygons (in a 2D domain) 1 Study on Delaunay tessellations 2 Figure 1: The seven final elements of the Ω Mesh Generator: (a) Cuboid, (b) Triangular Prism, (c) Quadrilateral Pyramid, (d) Tetrahedron, (e) Tetrahedron Complement, (f) Deformed Prism, and (g) Deformed Tetrahedron Complement. or polyhedra (in a 3D domain), that satisfy the Delaunay condition: the cir- cumcircle in 2D, or circumsphere in 3D, of each element does not contain any other mesh point in its interior [2]. The Delaunay condition is required because we use its dual structure, the Voronoi diagram, to model the control volumes in order to compute an approximated solutions. The basic elements used so far are triangles and quadrilaterals in 2D, and tetrahedra, cuboids, prisms and pyramids in 3D. Meshes composed of different elements types are called mixed element meshes [3]. Mixed element meshes are built on 2D or 3D domains described by sets of points, polygons or polyhedra depending on the application. We have developed a mixed element mesh generator [4] based on an extension of octrees [5, 6]. Our approach starts enclosing the domain in the smallest bounding box (cuboid). Second, this cuboid is continuously refined, at any edge position, by using the geometry information of the domain. That is why this refinement is called inter- section based approach. Once this step finishes, an initial non-conforming mesh composed of tetrahedra, pyramids, prisms, and cuboids is generated that fits the domain geometry. Third, these elements are further refined by bisection, as far as possible, until the density requirements are fulfilled. Fourth, the mesh is done 1-irregular by allowing only one Steiner point on each edge. The current solution is based on patterns but only the most frequently used patterns are available. Then, if a pattern is not available or the element can not be properly tessellated for the finite volume method, new Steiner points are inserted until all 1-irregular elements can be properly tessellated. The current set of seven final elements is shown in Figure 1. The advantage of using a mixed mesh in comparison with a tetrahedral mesh is that the use of different element types reduce the amount of edges, faces and elements in the final mesh. For example, we do not need to divide a cuboid into tetrahedra. On the other hand, a disadvantage is that the equations must be Study on Delaunay tessellations 3 Figure 2: (a) Mixed mesh of a 1-irregular cuboid that satisfies the Delaunay condition, (b) the same mixed mesh and its associated Voronoi diagram. discretized using different elements. Octree based approaches naturally produces co-spherical point sets. A mixed mesh satisfying the Delaunay condition can include all produced co-spherical elements as shown in Figure 2. The final elements in this example are five pyra- mids and four tetrahedra. The goal of this paper is to study the co-spherical elements that can appear while tessellating 1-irregular cuboids generated by using a bisection and inter- section based approach and to analyze how useful would be to include the new elements in the final element set. In particular, this paper gives the number and shape of new co-spherical elements needed to tessellate (a) all 1-irregular cuboids generated by the bisection approach and (b) some particular 1-irregular cuboids generated by the intersection refinement approach. In addition, statis- tics associated with particular tessellations are presented such as the frequency each co-spherical element is used and the number of tessellations that can be used with the finite volume method without adding extra vertices. The analy- sis of the 1-irregular cuboid tessellations was done under different criteria that affect the amount of generated co-spherical elements. We have focused this work on the analysis of the tessellations of 1-irregular cuboids because this element is the one that more frequently appears when meshes are generated by a modified octree approach. A theoretical study on the number of different 1-irregular cuboid configurations that can appear either by using a bisection or an intersection based approach was published in [7]. We use the results of that work as starting point for this study. This paper is organized as follows: Section 2 describes the bisection and intersec- tion refinement approaches. Section 3 presents briefly the developed algorithm to compute Delaunay tessellations. Section 4 and Section 5 give the results obtained by applying the algorithm to 1-irregular cuboids generated by a bisec- tion and an intersection based approach, respectively. Section 6 includes our conclusions. Study on Delaunay tessellations 4 Figure 3: Cuboid and its splits into two, four and eight cuboids using a bisection based approach. Figure 4: The bisection-refined cuboid at the left produces an 1-irregular ele- ment like the cuboid at the right . 2 Basic concepts This section describes some ideas in order to understand how 1-irregular cuboids are generated. 2.1 Bisection based approach 1-irregular configurations In this approach, the Steiner points inserted at the refinement phase are always located at the edge midpoints. Cuboids can be refined into two, four or eight smaller cuboids as shown in Figure 3. This refinement produces neighboring cuboids with Steiner points located at the edge midpoints. Those 1-irregular cuboids are larger than the already refined neighbor cuboid as shown in Figure 4. 2.2 Intersection based approach 1-irregular configurations While using an intersection based approach, the Steiner points are not neces- sarily located at the edge midpoints. In general, there are no constrains on the location of the Steiner points, except by the fact that parallel edges must be divided in the same relative position to ensure the generation of cuboids and not any other polyhedron. Figure 5 shows an example of this approach. This refinement produces neighboring cuboids with Steiner points located at any edge position. Those 1-irregular cuboids are larger than the already refined neighbor cuboid as shown in Figure 4. Study on Delaunay tessellations 5 Figure 5: Cuboid and its splits into two, four and eight cuboids using an inter- section based approach. Figure 6: The intersection-refined cuboid to the left produces an 1-irregular element like the cuboid to the right. 3 Algorithm In order to count the number of new co-spherical elements than can appear and to recognize their shape, we have developed an algorithm that executes the followings steps: 1. Build the point configuration of a 1-irregular cuboid by specifying the coordinates of the cuboid vertices and its Steiner points. 2. Build a Delaunay tetrahedral mesh for this point configuration by using QHull [8]1. 3. Join tetrahedra to form the largest possible co-spherical elements. 4. Identify each final co-spherical polyhedron. Qhull divides co-spherical point configurations into a set of tetrahedra by adding an artificial point that is not part of the input. Then, we use this fact to recognize the faces that form a co-spherical polyhedron and later to recognize which element is. 4 Results: Bisection based approach This section describes the results obtained by applying the previous algorithm to the 4096 (212) 1-irregular configurations that can be generated using a bisection based approach. First, the new co-spherical elements are shown. Then, their impact in all the tessellations is analyzed and finally, the tessellations that can be used with the finite volume method are characterized.